Mechanical arm imitation learning method based on meta-action hierarchical generalization

Abstract

A mechanical arm imitation learning method based on meta-action hierarchical generalization comprises the steps that an expert teaching data set is obtained and expressed as phi (s, a), s represents state data of all joints of a mechanical arm and comprises the spatial pose, the torque and the direction angle, and a represents the mechanical arm action mapped by the current state; the expert teaching data set is decomposed in a clustering mode, and a decomposed meta-action set is obtained; calculating the weight {pi1, pi2,..., pik} of each element action in the whole teaching action, wherein the weight coefficient pik is in one-to-one correspondence with the element action; according to the weight coefficient of each element action, generalization is carried out according to different proportions, and a generalization action is generated; sub-actions are randomly selected from the generalized meta-actions tau (A1, A2,..., Ak) to be combined, generalization teaching psi is obtained, the sub-actions are arranged according to expert teaching, and target actions identical to original expert teaching are formed; and inputting the generalization teaching psi into a convolutional neural network for supervised learning to obtain an execution strategy of the target action. According to the invention, the training efficiency and the accuracy of meta-actions are improved.

Description

Technical field [0001] The invention belongs to the field of robotic arm imitation learning, and specifically relates to a robotic arm imitation learning method based on hierarchical generalization of meta-actions. Background technique [0002] With the increasing use of intelligent robots and robotic arms in fields such as medical assistance, industrial manufacturing, express sorting, etc., although the traditional teaching programming method can achieve imitation learning and output of simple actions in specific scenarios, its learning The process requires the acquisition of a large number of manual teaching data sets, resulting in a large amount of manual labor redundancy; at the same time, the stability and robustness of the system working under continuous action need to be improved. In the invention patent CN 111983922 A, Lei Qujiang, Li Xiuhao and others from the Institute of Advanced Technology, Chinese Academy of Sciences in Guangzhou disclosed a robot demonstration ...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that imitation learning requires a large amount of expert teaching data in the strategy training process, and repeated manual teaching requires a lot of labor costs, and the cost of expert teaching is too high in some special environments, the present invention provides a method based on meta-action analysis. layer generalization method

Images